Cascading failure assessment of complex systems based on Bayesian networks

Abstract

A cascading failure assessment method based on Bayesian network (BN) is proposed in order to improve the reliability of complex system and show cascading failure longitudinal relationship among system, subsystems and components. The probability index of cascading failure is given by using conditional probability of BN which is transformed from fault tree (FT). After that, junction tree inference algorithm is adopted here to carry out bidirectional reasoning to exhibit quantitative assessment of influence on system failure due to subsystem or component failure and possibility of component failure under the condition of system failure. Finally, the method is applied to cascading failure assessment of 2-bus automatic alarm subsystem in ship wet sprinkler system to demonstrate its effectiveness.